hitsware said:
Well if it feels good, do it. 🙂
Personally, knob position has never been an issue for me. With the possible exception of instances where there's so much gain that knob positioning is excessively limited and I find it hard to dial in a "just right" setting.
se
What are exactly the problems that GNFB intoduces?
Unstability due to phase shift (neg fb becoming pos fb) is obvious
Mr Curl told about TIM and IMD poroblems, could you developp?
Is there anything else?
Unstability due to phase shift (neg fb becoming pos fb) is obvious
Mr Curl told about TIM and IMD poroblems, could you developp?
Is there anything else?
Hi,
This is all true and specifically true for the Williamson circuit which is, by design, on the brink of stability.
Hugh's analysis is quite correct although, IMO, it leaves out some important factors.
A triode has build in feedback by nature. How else would one be able to create a device with reasonable gain and still maintain some linearity?
This, however is local feedback, not global, and as such can react much quickly timewise.
The point of feedback encompassing the OPT is very real indeed and as a matter of fact OTL amps can tolerate much higher levels of GNFB which are much needed to reduce the Zout of the amp in order to be able to interface correctly with a common speaker's impedance.
IOWs, it wouldn't work without it and the same is true for most common sand amps.
By ommitting the OPT you can use more NFB but it doesn't have to include the input stages.
For tube amps using OPTs it would be advisable to use local NFB and distortion cancellation techniques, something that's certainly not all that easy to implement but it can be done.
One way is the use of cathode feedback winding(s) on the OPT but there are other ways.
GFNB is always running behind on what it's trying to correct; it's a kind of "Back to the Future" artefact.
It may seem to work statically but dynamically it sucks big time.
Same goes for most measurements made that are not dynamic, they don't reveal what's going on in "real time".
Cheers, 😉
This is all true and specifically true for the Williamson circuit which is, by design, on the brink of stability.
Hugh's analysis is quite correct although, IMO, it leaves out some important factors.
A triode has build in feedback by nature. How else would one be able to create a device with reasonable gain and still maintain some linearity?
This, however is local feedback, not global, and as such can react much quickly timewise.
The point of feedback encompassing the OPT is very real indeed and as a matter of fact OTL amps can tolerate much higher levels of GNFB which are much needed to reduce the Zout of the amp in order to be able to interface correctly with a common speaker's impedance.
IOWs, it wouldn't work without it and the same is true for most common sand amps.
By ommitting the OPT you can use more NFB but it doesn't have to include the input stages.
For tube amps using OPTs it would be advisable to use local NFB and distortion cancellation techniques, something that's certainly not all that easy to implement but it can be done.
One way is the use of cathode feedback winding(s) on the OPT but there are other ways.
GFNB is always running behind on what it's trying to correct; it's a kind of "Back to the Future" artefact.
It may seem to work statically but dynamically it sucks big time.
Same goes for most measurements made that are not dynamic, they don't reveal what's going on in "real time".
Cheers, 😉
>GFNB is always running behind on what it's trying to correct; it's a kind of "Back to the Future" artefact.
Yea that is a BASIC flaw (though in practice seems not to show much). Same thing as 'bootstrapping' (in reverse). The transit time of the signal through the amp makes either sorta nebulous.
Only really shows up (on my humble equipment) in recovery or settling time, but has audible implications beyond that.....mike
Yea that is a BASIC flaw (though in practice seems not to show much). Same thing as 'bootstrapping' (in reverse). The transit time of the signal through the amp makes either sorta nebulous.
Only really shows up (on my humble equipment) in recovery or settling time, but has audible implications beyond that.....mike
Hi,
The faster the response time of the loudspeaker the more it becomes obvious...
Take FR speakers with low moving mass, ELSs or ribbons and you'll quite likely hear the effect.
Most other, more conventional speakers, are way too slow.
Hope this helps,😉
The faster the response time of the loudspeaker the more it becomes obvious...
Take FR speakers with low moving mass, ELSs or ribbons and you'll quite likely hear the effect.
Most other, more conventional speakers, are way too slow.
Hope this helps,😉
>Believe me, if tube designers could use 60dB of NFB like many SS amps, they would!
Good one ! Thanks 🙂
You use single differential right?
Your amps are well regarded on
the FR driver forum..........mike
Good one ! Thanks 🙂
You use single differential right?
Your amps are well regarded on
the FR driver forum..........mike
hitsware said:
I've some serious reservations about that article.
By Ohm's law, the current in a speaker voice coil is the amplifier output voltage divided by the speaker impedance. Therefore, the current can only be linearly related to the voltage if the speaker presents a purely resistive load to the amplifier.
Excuse me? Reactive circuits violate Ohm's Law? Last I looked, current is linearly related to voltage in reactive circuits with only a relative phase change between current and voltage.
When, as is usually the case, the speaker impedance varies with frequency, amplitude, voice coil temperature etc., these dynamic impedance fluctuations translate directly into distortion of the driving current, as I = V / Z.
I = V/Z looks like a linear relationship to me. So where does he get this notion that current can only be linearly related to voltage if the load is purely resistive?
The negative feedback often used in power amps is totally blind to all of these problems, and with the reliance upon "damping factor" (electromagnetic-mechanic damping at resonance), huge back-EMF currents (at high voice coil velocities near resonance) are dumped back into the output of the power amp...
Huge back-EMF currents? Back EMF is greatest at and near resonance, which is where impedance is highest and by Ohms Law, current is lowest. So where are these "huge back-EMF currents" coming from? Hell, many loudspeakers' impedances exceed 100 ohms near resonance. There will be FAR more current in the circuit at the impedance minimums above and below resonance.
What's this guy talking about, Mike?
se
If we could just vote him off the island.........
"If I'm just trying to get attention, then you and Curl with your remarks such as these are just playing into my hands and drawing even more attention to me.
Who's looking for attention here? These comments, instead of simply going on about what it is you wish to discuss, seem to be saying "Hey, don't look at Steve, look at me!"
Go start your own thread then. I can't that any of your post have anything to do with differential input stages, or can't you get any response to your post accept by crashing threads where you have no knowledge of the topic? Be a man and stand on your own two feet instead of clinging to people's coattails till they get tired of you dragging things to a halt. I guess I could respond to people via Email instead of dealing with the constant threadjacking......... most of us are sick and tired your thread wrecking. I (and many others) are extremely sick of you ****ing Mr. Curl off in the middle of his giving extremely useful input on a topic. When you run somebody of his caliber off you are going to be even more unpopular than you are now.
Wake up and smell the coffee!
"If I'm just trying to get attention, then you and Curl with your remarks such as these are just playing into my hands and drawing even more attention to me.
Who's looking for attention here? These comments, instead of simply going on about what it is you wish to discuss, seem to be saying "Hey, don't look at Steve, look at me!"
Go start your own thread then. I can't that any of your post have anything to do with differential input stages, or can't you get any response to your post accept by crashing threads where you have no knowledge of the topic? Be a man and stand on your own two feet instead of clinging to people's coattails till they get tired of you dragging things to a halt. I guess I could respond to people via Email instead of dealing with the constant threadjacking......... most of us are sick and tired your thread wrecking. I (and many others) are extremely sick of you ****ing Mr. Curl off in the middle of his giving extremely useful input on a topic. When you run somebody of his caliber off you are going to be even more unpopular than you are now.
Wake up and smell the coffee!
I=V/Z is a linear relationship BUT
the Z of a driver (or any reactance)
is (by definition) not. Something to
do with 'rate of change' ........mike
the Z of a driver (or any reactance)
is (by definition) not. Something to
do with 'rate of change' ........mike
Hi,
Seems to me some people have difficulty in differentiating DC from AC behaviour...
Back to the kindergarten and Thevenin for some of us I reckon??
Please, people, do grow up and leave your inflated egos behind...
Cheers,😉
Seems to me some people have difficulty in differentiating DC from AC behaviour...
Back to the kindergarten and Thevenin for some of us I reckon??
Please, people, do grow up and leave your inflated egos behind...
Cheers,😉
Fred,
Not long back, when Steve so exasperated me with his behavior, at his suggestion I hit the 'ignore' button. Now I only look at his comments when I'm feeling benign, composed and amused. Works for me! 😀
Hitsware, the problem with feedback is exactly as Frank explains it; thanks Frank, you covered the gaps.
Here's another view, based on the order of events: input signal enters the amp, is magnified to the output (with errors), fed back to the feedback node and found to differ from the input. The error thus identified (by a diff pair with SS, with a tube it's usually cathode injection), creates an error signal which is despatched to the amp to CORRECT the output.
Notice the sequence of events. The signal takes time to pass through the amplifier, called group delay, and the correction is made AFTER the event since it's taken from the output and fed back to the input.
Now, in practice this is quick, particularly if fast devices (with high Ft) are used. So, a tiny error will soon be picked up, and corrected BEFORE it gets big. However, this creates overshoot, rather like a governor on a stationary diesel engine, which also reacts after the event, or temperature control of the human body, or respiration rate, triggered by carbon dioxide levels in the blood.
All feedback systems, absolutely by definition, will involve this delay, sometimes called hysteresis, which leads to tiny overshoot.
As it happens, by Fourier analysis, it can be shown that sharp blips in audio waveforms are made up of higher harmonics. And the problem is that even the initial overshoot will cause the feedback loop to react further, trying to correct the overshoot which results of the mechanism. In turn, this creates a cascading series of tiny overshoots, which spawn other overshoots, and so on. It is this unhappy chain of events which leads to the condemnation of global negative feedback, and like most folklore, there is some truth in it.
It would seem intuitive that if we knew the error the amp would introduce in advance of processing the signal, we could make alterations to our amplifying devices so that it came out just right!! In truth, we do this to a minor degree by ensuring that each gain block is as linear as possible to begin with. This minimizes the correction necessary using global feedback, but a moments thought tells us that to achieve this we will need to modify the transfer function of our active devices to make them more linear. And guess what! To do this we reduce the gain of our devices on the spot (locally), but this actually reduces the feedback factor we can apply globally!! SNOOKERED!!
You really are damned if you do, and damned if you don't. IOW, amplifier design is just life - these sorts of dilemmas are a part of life itself. The classic expression of this, in macabre but entirely logical terms, is 'The operation was a success, but the patient died'.
One valid, real-world approach, in my opinion, is to have extremely fast devices and very short, non-inductive feedback and signal paths which minimize group delay and thus bring down overshoot and distortion to vanishingly low levels. We should also use very fast topologies, where possible, and make the slowest part of the amplifier the gain block, so as to control phase shift to avoid oscillation. Since the ear is preternaturally sensitive to high order harmonics, particularly odds, we have to design amplifiers which bring the H5 and beyond harmonics below the noise floor, so that any sonic impact of this unfortunate but necessary topology is subliminal rather than directly audible. Even so, no amplifier is perfect, and the discussion of sonic color added by the SET brigade brings up a whole new can of worms. I'd recommend reading Fred Nachbaur's website here: http://dogstar.dantimax.dk/tubestuf/miniblok.htm for a full and intelligent discussion of the SET, which incidentally I like very much.
You ask about the AKSA. Yes, it is a single differential input pair, using highly matched transistors, very carefully chosen, and VERY carefully laid out.
Cheers,
Hugh
Not long back, when Steve so exasperated me with his behavior, at his suggestion I hit the 'ignore' button. Now I only look at his comments when I'm feeling benign, composed and amused. Works for me! 😀
Hitsware, the problem with feedback is exactly as Frank explains it; thanks Frank, you covered the gaps.
Here's another view, based on the order of events: input signal enters the amp, is magnified to the output (with errors), fed back to the feedback node and found to differ from the input. The error thus identified (by a diff pair with SS, with a tube it's usually cathode injection), creates an error signal which is despatched to the amp to CORRECT the output.
Notice the sequence of events. The signal takes time to pass through the amplifier, called group delay, and the correction is made AFTER the event since it's taken from the output and fed back to the input.
Now, in practice this is quick, particularly if fast devices (with high Ft) are used. So, a tiny error will soon be picked up, and corrected BEFORE it gets big. However, this creates overshoot, rather like a governor on a stationary diesel engine, which also reacts after the event, or temperature control of the human body, or respiration rate, triggered by carbon dioxide levels in the blood.
All feedback systems, absolutely by definition, will involve this delay, sometimes called hysteresis, which leads to tiny overshoot.
As it happens, by Fourier analysis, it can be shown that sharp blips in audio waveforms are made up of higher harmonics. And the problem is that even the initial overshoot will cause the feedback loop to react further, trying to correct the overshoot which results of the mechanism. In turn, this creates a cascading series of tiny overshoots, which spawn other overshoots, and so on. It is this unhappy chain of events which leads to the condemnation of global negative feedback, and like most folklore, there is some truth in it.
It would seem intuitive that if we knew the error the amp would introduce in advance of processing the signal, we could make alterations to our amplifying devices so that it came out just right!! In truth, we do this to a minor degree by ensuring that each gain block is as linear as possible to begin with. This minimizes the correction necessary using global feedback, but a moments thought tells us that to achieve this we will need to modify the transfer function of our active devices to make them more linear. And guess what! To do this we reduce the gain of our devices on the spot (locally), but this actually reduces the feedback factor we can apply globally!! SNOOKERED!!
You really are damned if you do, and damned if you don't. IOW, amplifier design is just life - these sorts of dilemmas are a part of life itself. The classic expression of this, in macabre but entirely logical terms, is 'The operation was a success, but the patient died'.
One valid, real-world approach, in my opinion, is to have extremely fast devices and very short, non-inductive feedback and signal paths which minimize group delay and thus bring down overshoot and distortion to vanishingly low levels. We should also use very fast topologies, where possible, and make the slowest part of the amplifier the gain block, so as to control phase shift to avoid oscillation. Since the ear is preternaturally sensitive to high order harmonics, particularly odds, we have to design amplifiers which bring the H5 and beyond harmonics below the noise floor, so that any sonic impact of this unfortunate but necessary topology is subliminal rather than directly audible. Even so, no amplifier is perfect, and the discussion of sonic color added by the SET brigade brings up a whole new can of worms. I'd recommend reading Fred Nachbaur's website here: http://dogstar.dantimax.dk/tubestuf/miniblok.htm for a full and intelligent discussion of the SET, which incidentally I like very much.
You ask about the AKSA. Yes, it is a single differential input pair, using highly matched transistors, very carefully chosen, and VERY carefully laid out.
Cheers,
Hugh
Hugh,
Thanks for the link. 🙂
Could we say (getting back to single vs dual differential)
that perhaps the single differential has a little of the euphoric
qualities of the SET while the dual doesn't ? Or does the
' differentialness ' destroy that as compared to say a single jfet or bipolar xsistor as a front end. ( I know there is some differance between the 2 (fet vs bipolar) ......... Thanks again ...... mike
Thanks for the link. 🙂
Could we say (getting back to single vs dual differential)
that perhaps the single differential has a little of the euphoric
qualities of the SET while the dual doesn't ? Or does the
' differentialness ' destroy that as compared to say a single jfet or bipolar xsistor as a front end. ( I know there is some differance between the 2 (fet vs bipolar) ......... Thanks again ...... mike
Hi,
Magnificent post, Hugh.
This reminds me of the only possible flaw I made when designing my OTL amps...I should have made the feedback path shorter by building along the width of the chassis not the length.
Thank you for reminding me,😉
Magnificent post, Hugh.
This reminds me of the only possible flaw I made when designing my OTL amps...I should have made the feedback path shorter by building along the width of the chassis not the length.
Thank you for reminding me,😉
Let's keep them both : )
From the peanut gallery, I certainly value John Curl's presence at a public forum; at the same time, I also value Steve Eddy's opinions. My wish is that they would give each at least another chance. Seems like they are both close to, if not on the mark. Of course, everyone has his or her quarks. Just don't let the other bother you. Seems like the same goal is in mind.
So, metaphorically, I'm for dual differential (more local feedback, less global feedback).
JF
From the peanut gallery, I certainly value John Curl's presence at a public forum; at the same time, I also value Steve Eddy's opinions. My wish is that they would give each at least another chance. Seems like they are both close to, if not on the mark. Of course, everyone has his or her quarks. Just don't let the other bother you. Seems like the same goal is in mind.
So, metaphorically, I'm for dual differential (more local feedback, less global feedback).
JF
Maybe the that Japanese design is trying to search the limit of minimum power amp stages. But I think the best 2 stage power amp is Aleph. The 2 stage power amp with only 1 differential. I've tried to figure 2 stage power amp, and never got better idea than Aleph. Good sound, good stability (although very hot).
Reading the previous post, I think I began to understand what Mr.Pass said about "Anchoring" the whole amp with CCS. Dual differential cannot do that, the standing current will vary as the transistor getting hotter or more cold (bias in VBE to temp). If stability of standing current is important to output signal for an audio amp, then single differential with CCS is the best. Although it is the "OLD DESIGN" like some refer to Doug Self design.
But IF single differetial with CCS is better from dual differential, why is there so many famous brand (like Crest) is using dual differential up until now?
Reading the previous post, I think I began to understand what Mr.Pass said about "Anchoring" the whole amp with CCS. Dual differential cannot do that, the standing current will vary as the transistor getting hotter or more cold (bias in VBE to temp). If stability of standing current is important to output signal for an audio amp, then single differential with CCS is the best. Although it is the "OLD DESIGN" like some refer to Doug Self design.
But IF single differetial with CCS is better from dual differential, why is there so many famous brand (like Crest) is using dual differential up until now?
Don't want to go off topic, but one: I was writing metaphorically and two (and my ducks aren't in a row yet): I'm interested in a cross-couple differential out design that uses only N-channel JFETs and NPN follower or only N-channel JFETs (for a headphone amplifier). Again, why? "Sounds" good to me. I planned to sketch this out for another thread. I wonder if the N-channel and NPN components tend to be better, why they aren't just used in a bridged mode (and it may be easier for a headphone amp). In a way, mirror the circuitry East and West (versus North and South). I'll take this to another thread.
JF
JF
To keep it simple:
Dual differential permits use of a dual voltage amplifier; completely symmetrical all around. This looks great and works well.
However, and accounting for the minor transfer function differences between NPN and PNP, their distortions largely cancel, and certainly this is true also in the voltage amplification sections. The result of this is that only symmetrically disposed distortions remain - the asymmetrical ones cancel pretty well - and you are left with very low levels of odd order.
The single differential input drives the designer towards use of a single-ended voltage amplifier. This creates considerable asymmetrical distortion, or even order harmonics, which is largely diminished by the global negative feedback loop (as are all distortions, regardless of order).
This generally means that dual diff will have lower levels of even order, and low levels of odd order. But single diff will have low levels of even and odd; all are produced, however, not just odds, and with reasonably fast GNFB the incidence of these harmonics will progressively lessen with order.
The SET sound has progressively lessening harmonics; petering out at around the 5th harmonic. With a well designed GNFB SS amplifier, it is possible to emulate this profile to a certain extent, though the levels of distortion will be typically two orders of magnitude superior to the SET, but with a few more distortion artefacts.
My own belief is that it is the recording process which strips away the higher harmonics of recorded music. The SET goes part the way to reconstituting this denuding of the harmonic profile, richening up the sound, literally 'filling' it out.
Cheers,
Hugh
Dual differential permits use of a dual voltage amplifier; completely symmetrical all around. This looks great and works well.
However, and accounting for the minor transfer function differences between NPN and PNP, their distortions largely cancel, and certainly this is true also in the voltage amplification sections. The result of this is that only symmetrically disposed distortions remain - the asymmetrical ones cancel pretty well - and you are left with very low levels of odd order.
The single differential input drives the designer towards use of a single-ended voltage amplifier. This creates considerable asymmetrical distortion, or even order harmonics, which is largely diminished by the global negative feedback loop (as are all distortions, regardless of order).
This generally means that dual diff will have lower levels of even order, and low levels of odd order. But single diff will have low levels of even and odd; all are produced, however, not just odds, and with reasonably fast GNFB the incidence of these harmonics will progressively lessen with order.
The SET sound has progressively lessening harmonics; petering out at around the 5th harmonic. With a well designed GNFB SS amplifier, it is possible to emulate this profile to a certain extent, though the levels of distortion will be typically two orders of magnitude superior to the SET, but with a few more distortion artefacts.
My own belief is that it is the recording process which strips away the higher harmonics of recorded music. The SET goes part the way to reconstituting this denuding of the harmonic profile, richening up the sound, literally 'filling' it out.
Cheers,
Hugh
Re: If we could just vote him off the island.........
Why? Nothing I'd said was any more out of context with the original post in this thread than what a number of others were talking about.
I'm not the one who started talking about power amps, output stages and negative feedback. I just went with the flow of the discussion.
You mean you missed the differential input stage in that circuit I posted? How'd you miss that? It was the only other stage beside the output stage.
Well, John said:
SE is correct, if you want to make an input filter, then put 50-200pf mica or film cap to ground to have a LINEAR capacitance at the jfet input.
So I guess I know a little something.
What are you talking about? You don't have to deal with anything. You're not a moderator here. Either report my posts to those who are moderators, or put me on ignore.
What are you talking about? I didn't say a word to John and nothing I had posted prevented John from going about what he was doing.
Don't have to worry about that happening. He followed me over here. He tried to say it was so he could get away from me over at that other place but in fact he knew I was posting here so he came over here so he could **** and moan about me.
He's something of an obsession with me. He even goes onto other message boards that I don't participate on and ****es and moans about me.
So don't worry. John would be more likely to stop posting here if I left than if I stayed.
se
Fred Dieckmann said:
Why? Nothing I'd said was any more out of context with the original post in this thread than what a number of others were talking about.
I'm not the one who started talking about power amps, output stages and negative feedback. I just went with the flow of the discussion.
You mean you missed the differential input stage in that circuit I posted? How'd you miss that? It was the only other stage beside the output stage.
Well, John said:
SE is correct, if you want to make an input filter, then put 50-200pf mica or film cap to ground to have a LINEAR capacitance at the jfet input.
So I guess I know a little something.
What are you talking about? You don't have to deal with anything. You're not a moderator here. Either report my posts to those who are moderators, or put me on ignore.
What are you talking about? I didn't say a word to John and nothing I had posted prevented John from going about what he was doing.
Don't have to worry about that happening. He followed me over here. He tried to say it was so he could get away from me over at that other place but in fact he knew I was posting here so he came over here so he could **** and moan about me.
He's something of an obsession with me. He even goes onto other message boards that I don't participate on and ****es and moans about me.
So don't worry. John would be more likely to stop posting here if I left than if I stayed.
se
- Status
- Not open for further replies.